Physiology-Aware Gaussian Head Avatar Embeds Recoverable Cardiac Signals into Facial Appearance
Overview
Gaussian head avatars typically model intrinsic facial appearance as temporally static, omitting the subtle cardiac-induced skin-color variation that occurs naturally in real faces. In a 2026 paper accepted to SIGGRAPH Asia 2026 Technical Communications, researchers introduce Heartian, a physiology-aware modulation framework that learns cardiac-cycle-dependent, per-frame albedo modulation of facial skin-region Gaussians within a relightable head avatar, encoding remote photoplethysmography (rPPG) signals directly into the avatar's material attributes.
As rPPG-based physiological sensing matures in 2026 β increasingly used in telemedicine, driver monitoring, and affective computing β the ability to synthesize avatars that preserve realistic, recoverable cardiac signals has become relevant for benchmarking and privacy-aware synthetic data generation.
Methodology
Using synchronized contact PPG supervision, Heartian models the prescribed cardiac waveform as the sum of two Gaussian functions and learns per-frame spatial residuals via a lightweight MLP. This design allows cardiac-cycle information to be embedded as a controllable material attribute rather than baked into static appearance.
Results
Across 152 stationary recordings from UBFC-rPPG, PURE, and MMPD, attribute-space recovery of the supplied signal achieves a pooled recording-level heart-rate MAE of 0.29 bpm and MAPE of 0.38%.
Crucially, the signals remain detectable after rendering by benchmark rPPG methods. The best tested configuration β a motion-augmented TS-CAN decoder pretrained on UBFC-rPPG β recovers heart rate from the rendered MMPD avatars at 0.97 bpm MAE and 1.21% MAPE.
Meanwhile, Heartian maintains reconstruction quality comparable to the baseline, with a negligible average PSNR degradation of 0.005 dB.
Significance
The work demonstrates that recoverable rPPG signals can be embedded as controllable material attributes in subject-specific Gaussian head avatars while retaining reconstruction quality. This bridges physiological signal processing with relightable neural rendering β a connection with growing relevance in 2026 as Gaussian splatting-based avatars become standard in virtual production, telepresence, and synthetic media pipelines.
Publication Details
- Authors: Xiaoyue Fan, Jose Echevarria, Akshay Paruchuri, Kaan AkΕit
- Comments: 4 pages of manuscript and 2 pages of supplementary material; SIGGRAPH Asia 2026 Technical Communications
- Subjects: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR)
- Cite as: arXiv:2609.28539 [cs.CV]
- DOI: https://doi.org/10.48550/arXiv.2609.28539
- Submitted: 22 Sep 2026
via ArXiv CV